DeepSeek V4 has been fully adapted to Huawei Ascend 950 series chips

robot
Abstract generation in progress

Golden Finance reports that on April 24th, DeepSeek announced that the preview version of DeepSeek V4 is officially launched and open-sourced. The official statement from DeepSeek indicates that DeepSeek V4 Pro currently has very limited service throughput, and it is expected that in the second half of the year, the Ascend 950 supernodes will be mass-produced. At that time, the Pro price will drop significantly. Additionally, Huawei’s Ascend CANN Bilibili account posted that a live broadcast will be held today at 16:00, with the theme: DeepSeek V4 Ascend Launch.
It is understood that the Ascend 950 supernode refers to Huawei’s supercomputing unit built on the “supernode” architecture, equipped with Ascend 950 series AI chips. Its product name is Atlas 950 SuperPoD, which integrates thousands of Ascend 950 DT chips into a logical “giant AI computer” through high-speed interconnection technology, specifically designed for training and inference of trillion-parameter-level large models.
This indicates that DeepSeek V4 has fully adapted to Huawei’s Ascend 950 series chips, but lacks sufficient computing power.

View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pin